Well, it looks like there are some rather nasty lurking bugs in
gcc-2.95.1; bugs that popped up when things were compiled with -O2 went
away when compiled with -O; I'm beginning to suspect that the alias
analysis code is a bit stricter than a lot of programmers out there.
Anyway, I was getting crashes in code Theme.cpp in ParseColor, when
I inserted the extra lines (@845)
char *foo = oColorstring.c_str();
printf("colors: %s; ocolor: %lx", foo, &oColor);
And then the crashing stopped; unfortunately, the printf never showed
itself, and the crash has occured deeper in the code in GTK/glib land:
(gdb) bt
#0 0xff2fdd74 in XGetWindowAttributes () from /usr/openwin/lib/libX11.so.4
#1 0xfe3dcae8 in gdk_window_foreign_new (anid=0) at gdkwindow.c:508
#2 0xfe2ce478 in GTKBitmap::GTKBitmap (this=0x171198, oName=@0x99a9c)
at ui/theme/unix/src/GTKBitmap.cpp:59
#3 0xfe2a2294 in Theme::BeginElement (this=0x16d2c4, oElement=@0xffbe34f0,
oAttrMap=@0xffbe33f0) at ui/theme/src/Theme.cpp:296
#4 0x4ff94 in Parse::DoParse (this=0x16d2c4) at lib/xml/src/Parse.cpp:225
#5 0x4efcc in Parse::ParseFile (this=0x16d2c4, oFile=@0xffbe38d0)
at lib/xml/src/Parse.cpp:61
#6 0xfe2a0748 in Theme::LoadTheme (this=0x16d2c4, oFile=@0xffbe3b08)
at ui/theme/src/Theme.cpp:155
#7 0xfe2c0254 in FreeAmpTheme::LoadFreeAmpTheme (this=0x16d2c0)
at ui/theme/src/FreeAmpTheme.cpp:151
#8 0xfe2bfb10 in FreeAmpTheme::FreeAmpTheme (this=0x16d2c0, context=0x966f0)
at ui/theme/src/FreeAmpTheme.cpp:83
#9 0xfe2bf72c in Initialize (context=0x966f0)
at ui/theme/src/FreeAmpTheme.cpp:59
#10 0x3243c in Player::Run (this=0x98708) at base/src/player.cpp:645
#11 0x49508 in main (argc=709, argv=0xffbe3f44)
at base/unix/src/bootstrap.cpp:132
the last bit of code that's in freeamp proper is frame #3,
#3 0xfe2a2294 in Theme::BeginElement (this=0x16d2c4, oElement=@0xffbe34f0,
oAttrMap=@0xffbe33f0) at ui/theme/src/Theme.cpp:296
296 pBitmap = new GTKBitmap(oAttrMap["Name"]);
Which isn't inspiring confidence in me.
At this point, I'm thinking of trying the egcs-current source, and
seeing if those mollify any bugs; it looks like the bleeding-edge
template stuff is causing gcc (or me, for that matter) to not be a
Happy Boy. (The fact that linkage problems mysteriously go away
with -fsquangle do not fill me with joy, either.)
Does any of this crap show up in linux, particularly the x86 flavors?
